Attention Relational Graph Convolution Networks for Relation Prediction in Knowledge Graphs

Recently, the tasks based on knowledge graph (KG), such as question answering, information retrieval, are more and more widely used. However, due to the incompleteness of the relations (links) between entities, it is very important to study the relations prediction based on KG to complete the missing relations between entities. In recent years, graph convolution networks (GCNs) have been a new method to solve the reasoning of knowledge graph. However, the existing knowledge graph relation prediction model based on GCNs fails to consider the importance between nodes. In order to obtain more abundant relations information between entities (nodes), inspired by the graph attention network (GAT), we propose an attention weighted relational graph convolutional network (denoted as AWR-GCN), which is used as an encoder of the encoder-decoder model for relationship prediction, and the decoder is a linear model. Compared with the advanced methods in the commonly used relational prediction data sets, our model has a certain performance improvement and reached advanced level.